Adaptive Cell-Size HoG Based Object Tracking with Particle Filter

Visual object tracking is one of the most vigorous research area in Computer Vision. A lot of algorithms already achieve high performance and accuracy. Whereas, most of visual object tracking algorithms proceed separately from detection algorithm because of difference between tracking and detection descriptor. Instead, we propose adaptive cell-size HoG (acHoG) based Particle Filter Tracking (PFT) algorithm. Using HoG enables to share information with detection algorithm based on HoG and complements the fast but inaccurate PFT. However, according to the characteristics of PFT, features are extracted more than two times on the most of the target region. To solve repeated feature extraction problem, we apply adaptive cell-size to HoG. Because acHoG shares intensity and angle of edge extracted from original target image, repeated computations can be reduced and same size features extracted from different size targets. Experimental results have proved that the acHoG is effective. 540 Whui Kim and Kyong-Ho Kim